SpectroSzNet: Deep Learning-Based Enhancement for Schizophrenia Classification Using Spectrograms

Schizophrenia is a neuropsychiatric disorder characterized by delusions, auditory hallucinations, mood swings, and unusual behavior. Early detection of Schizophrenia can be helpful to cure the patient before the condition gets worse. However, the early diagnosis remains a formidable challenge because of the complex, diverse, and vague symptoms of the disorder. The need for a straightforward and reliable diagnostic approach to identify schizophrenia has led us to explore the potential of Electroencephalography (EEG)-based models. Auditory processing impairments have been linked to clinical symptoms and cognitive problems in individuals with schizophrenia and it can be studied using the EEG recordings of the patient. For the early diagnosis of schizophrenia, SpectroSzNet, a deep learning-based EEG-centric solution is introduced in this paper. SpectroSzNet aims to enhance the detection and classification of Schizophrenia using EEG data. The incorporation of both raw EEG data and spectrogram features in the SpectroSzNet model plays a pivotal role in achieving high accuracy in classification. For this task, the data from five midline channels out of all the EEG electrodes have been utilized. Results from the experiments show that the proposed model can classify with an accuracy of 87.02%.